AMD Ryzen: Good For Enthusiasts, Bad For Investors

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.
AMD really could have done better with the Ryzen marketing and product line.

1- Should have been called R8, R6, R4. It's better than i7, i5, and i3 because it has bigger numbers AND easier to understand. 8, 6 and 4 cores.
2- Drop the X models since all the models have XFR anyway. Once you drop the X models, market all models to have higher base boost since that's what XFR is anyway, 100MHz boost. (Non X models had 50)
3- 6 cores models should be using the 2 CCX with disable cores like it is now but the 4 core model should have been 1 CCX. Now you saved the cost of the additional CCX and increased performance at the same time.

Please hire me AMD. You need help.
 


1. Im sorry i'm going to have to disagree, when the other company has more swing, they dictate the name and you deal with it, when you have the unquestionably better product or superior market share, then you change it to what you want.

2. Yeah I dunno what they're thinking there either. Also it should just be the 1800x and 1700, the 1700x confuses things as someone else mentioned earlier. Also the 1800x should probably be price corrected to close to $420 so it can go down to $399 and be "on sale". To be fair they will probably make those adjustments, but when you can get more during the first few months, you milk what you can. They will probably wait for the 1900x or whatever comes out though, unless Intel does some unlikely crazy shit.

3. The reason there are no dedicated 4 core dies yet is that they simply aren't out yet and they're just harvesting what they can from their initial big batches of 8 cores. I see that changing when the apu's come out since they will probably be dedicated 4 core/8 thread designs (How much, if any L3 cache is the question though).

Honestly I'm more curious about what their lineup will look like 6 months from now, as things stand, it really has not been a bad launch.
 


You completely missed the point. AMD's gross margin guidance for the next quarter is lower than the previous quarter, meaning profit margins will be even lower when Ryzen is shipping for the whole quarter.
 

With AMD shipping more lower-end parts based on the same high-end die, the margin crunch is unavoidable. AMD must be betting on volume to make up the net difference. If AMD fails at doing so, things may get uglier still next quarter with the R3-1100/1200/1300.
 
Seriously Toms? You can review financial annual result with a 20 days windows now? Ryzen was released for 20 days or so during Q1 2017.

This is pure desinformation... and by the way, Goldman Sachs and another firm devaluated AMD for no reasons except their gut feelings.
 

The question is, are the higher tier SKUs actually higher binned chips? From what I've seen online about Ryzen, binning is minimal to non existent. No one's faulting them for attempting to sell some flagship chips at a bit of a premium. The problem is that the various price points for essentially identical performance makes those flagships very unattractive buys, causing AMD's attempt at product segmentation to fall apart at somewhat. For the consumer, it's great that you can get flagship performance with a ~$330 chip. For AMD's revenue and attempt at product segmentation, maybe not so much.
 
One of the things that analysts are missing is how Ryzen will perform in the server market. All the performance measures so far indicate they Ryzen does well in tasks that aren't gaming. As the WSJ recently reported, AMD has close to 0% of the server market. If they can capture just 20%, that will be a huge jump in revenue for the company even if margins aren't great.

The design cycle for boards is about 6 months so if the server chips are introduced this quarter, POs for chips won't be showing up till the 3rd or 4th quarter of the year. AMD is playing this for the long haul.
 
Tech stock analysts still predict the stock to be a 15$+ stock. With everything just around the corner, it's hard to pinpoint a 10$ value. After such a news, selling should still occur however the stock is fighting to keep it's up trend of the last year.

Toms, leave financial analyses to analysts.
 


I agree on if AMD holds 20% of the market share is a huge win for them in compere to previous years and I think this is what they had a goal in the beginning.


The problem is not in the AMD CPU that are not good for gaming, the problem comes from the ignorance of the community or who get tricked by all these benchmarks out there. For example, The 7700 gives you 150FPS for game X while 1600X gives you 100FPS for the same game but costs less. What these FPS numbers mean to me if I have a monitor that is 60 refresh rate? Absolutely Nothing! As both will cap the monitor FPS which is 60 and what is left to me is to take cost in consideration.
 
I think it all seems like a massive over reaction.

1. Ryzen was never meant to be a high margin product. It was meant to win back the hearts and minds of "enthusiasts" and bring some competition to the mid-range and high-end desktop segment. This "shock therapy" came at the cost of margins, as all models are unlocked and based on the same die. AMD needed to do this, because mind share is important, and AMD had lost all credibility in this segment.

2. Mobile APU:s based on Zen are coming. True 4-core dies are coming. No need to freak out. Yeah, it's a shame we won't get a Zen 1 based desktop APU, but it's simply not as important as mobile these days, and AMD had to prioritize. Besides, they still have Bristol Ridge, which is actually pretty decent for its price point and works with the same, up-to-date AM4 platform as Ryzen.

3. Naples. Gaining market share in the server space where the margins are much higher would be absolutely huge for AMD. Zen is clearly optimized for this, perhaps at the expense of clock speeds and gaming performance. 64-thread Naples chips running at 2.x GHz are going to blow Intel away in terms of power and price/performance, making them extremely attractive for cloud service provides and other large-scale server operators.
 
I thought the final verdict (after release) was that the 1400/1500's were only one CCX after initial data said it was going to be 2+2 double CCX config...
 

Compared to AMD's gross margins on the CPUs it offered for the past 4-5 years, margins on Ryzen should be a whole lot better - AMD's cheapest Ryzen currently retails for more than their most expensive previous generation best offer.

Having so much lost ground and market share to claw back does limit AMD's ability to inflate prices. The many platform and performance quirks that are still being worked on don't help either.
 
It's been very good for me as an investor. It's up 300% for since I bought it. Wish bought more at the bottom for a 700% return.

At the time I thought I should buy AMD stock instead of a new rig while waiting for the next big thing to be released. Then I could sell AMD to buy AMD.
 


"..The "X" and non-"X" processors feature similar overclocking headroom.." from the article
"... From what I've seen online about Ryzen, binning is minimal to non existent. ..." from above post

The prices make a bit more sense if you assume that higher priced chips have more headroom.

the 1500X article seems to support binning working (and undermine the "same headroom" comment)

"We spent a considerable amount of time tuning the Ryzen 5 1500X to match our previous efforts with Ryzen 7- and 5-series processors. Eventually, we did dial in a Prime95-stable 3.9 GHz overclock at 1.375V and an auto LLC (Load Line Calibration) setting. In the U.S. lab, we recorded up to 64°C (per AIDA) with our Noctua NH-U12S SE-AM4 during extended stress tests. Seeing plenty of available thermal headroom, we then attempted to match the 4 GHz we saw from other Ryzen processors (except the 1700), but were unsuccessful."

"Finally, we got 2933 MT/s running stably, but that appears to be our sample's ceiling. Interestingly, we achieved 3200 MT/s with other Ryzen processors on the same motherboards, implying the disparity stems from the chip's IMC (Integrated Memory Controller)."
 

That'a a valid point for sure. 3.9-4Ghz seems a pretty hard ceiling with only the rare CPU stable at 4.1 with voltages acceptable for 24/7 usage. So maybe there's just not a lot of CPU to CPU variation which makes the whole binning process less useful.

While I agree that AMD's market segmentation doesn't really work in that there's not enough (or any?) incentive to pay for higher SKUs, I think the issue is that if they wanted to compete outside of pure workstations, they simply didn't have any other choice.
Intel can use the same die for their $1700K 10 core and $390 6 core because all are (at least somewhat) competitive at their price points.
Similarly they can use the same die for their $340 7700K and the $180 7500 because again - all are competitive at their price points.

AMD don't have that luxury. Unless they want to reduce Ryzen to a workstation CPU and abandon the gaming market altogether, they need a low $200 6 core and low $300 8 core to be competitive. And they need those SKUs to hit high 3Ghz clocks to be competitive... so what other option do they have? If you need 90-95% of the peak performance of your product to compete at ~$330, then your options for higher margin and compelling CPUs at $400- $500 are limited.

If (hypothetically speaking), the best Ryzen CPUs could get 5Ghz, then cut-down Ryzen CPU with locked multipliers and 4 and 6 cores could command higher prices, then perhaps the unlocked 4 core could compete at over $200, and the 6 cores compete at over $300, allowing them to spread out their products and thus justify a fully unlocked and highly clocked 8 core at $500... they'd have solid market segmentation. But unfortunately Ryzen's performance (while great IMHO), wasn't strong enough to provide competition and compelling products across that entire $150 -> $500 price range. So they did what they had to do, identify a couple of key price/performance targets and nailed them with the Ryzen 5 1600 and Ryzen 7 1700... then filled out the rest of the price brackets as best they could (which - I totally agree - don't make much sense).

So I agree that their market segmentation doesn't really work, I just don't think there was any other way to chase meaningful market share with the product they had.
 
I think its ignoring the fact that amd is able to use die modules with two defective cores to make there 4 core cpus. Other wise it would just be cost more money in the trash then just the higher cost (it inproves the usefull chip yields). Yes it does still create a power and speed hit. Better to make $10 on using two two core chips then trowing both in a dumpster at a cost of $30 (incomplete chips dont cost the full $50 package cost)
 

And people who say what you are saying are ignoring the facts that AMD could make about twice as many quad-core dies per wafer and that if defect density rate was so high that AMD had a significant number of dies with two bad CPUs per CCX, there is no way in heck that the R5-1400 would be the only model with 8MB of L3 cache. L3 cache accounts for half of the CCX die area and any defect in the L3 cache means that the die gets binned as R5-1400 or worse, which means you are about four times as likely to get defective L3 as you are of getting a single bad core on a given CCX.

AMD's yields would need to be horrific for this to make sense.
 
Keep in mind the computer trading. if a stock falls to a certain price, a sell order is placed automatic which activate another sell order causing a selling off of a stock. The stock to my way of thinking is very undervalued.
 

R3 will likely have 8MB cache models as well. Plus you keep repeating this idea that the second CCX is practically half the die. It's not true. They could not make "about" double the chips with a single CCX even if you really want to argue over the all-encompassing stretchy word "about". A single CCX + rest of the chip = what, 2/3 the size of a dual CCX chip? A little smaller? Plus it means you have to either throw out single-CCX chips with defective cores or bin down to tri- and dual-cores. Not to mention the time/cost/engineering resources would be taken away from Naples and Raven Ridge.

But yes, clearly all the armchair engineers know best.
 

Maybe I'm missing something here, but I don't see how that example can be considered evidence of binning. The 1500X is the highest tier (and therefore supposedly highest binned) quad core Ryzen chip, and yet they experienced ho-hum CPU/memory overclocking. I see that as being indicative of a lack of binning, not the other way around. When I referred to binning, I specifically meant chips with the same number of cores, where the only thing to differentiate them is frequency. Given that all chips are unlocked, stock frequency doesn't matter much, and that leaves overclocking headroom to differentiate the chips, which it fails to do.

Edit: OK, the 1500X isn't the best example, as it does actually have something other than clock speed to differentiate it from the 1400: double the L3 cache. But what I said still applies to the 6/8 core Ryzen CPUs.
 

I'm sure they're binning... you can bin for power/voltage too. There's clearly some issues holding Ryzen's clocks back, so their binning for clocks is either minimal, or simply unsuccessful at this point. But binning for TDPs is still binning, so it's not like they're not doing anything at all.

They definitely need to work on better revisions, once they finish their rollout. They have a lot on their plate, though... Naples and mobile APUs are both top priority.
 
Hype had the Ryzen processor as an Intel killer akin to the first AthlonX64 chips. It isn't. The market had bet on that assumption and now it is simply correcting itself. That is why being a publicly traded company is so dependent on public perception. I think that the Ryzen chip is a great product, just two to three years to late to market. The investors are also nervous about vega. With Nvidias recent moves with 1080ti and Titan Xp and volta in the near future AMD is continuing to put out innovative products that are just a little to little to late. Low balling price in the tech game is a time proven strategy to go broke. If you want to be in the tech business you need to be on top of the performance curve pretty regularly or you risk a total collapse in your ability to demand strong profit margins. This takes strong R&D budgets, good people and ultimately margins to fund both. The "good enough" mainstream processor/gpu isn't where the money is at.
 

I never said half the die, I said 35%, which is 1/3 and includes chunks that appear duplicated for each CCX. Once you throw out the die-stitching port which uses 5% of the die space (likely more after axing the associated paths through the fabric and related structures) and throw out support circuitry for the second CCX, the remaining IOs (DRAM, PCIe, USB and unidentifiable stuff along the right edge) only use 20% of the die. Re-arrange the die to reduce under-used space (the dual-CCX layout looks horrible in terms of space efficiency thanks to its wide aspect ratio) and a single-CCX chip at ~55% of the die size is within feasible territory.

As for the time/cost engineering, having the ability to efficiently spin off chips tailored to specific markets at a is the whole point of modular designs. If adding or eliminating a stop on the fabric requires more than relatively trivial effort that hasn't already been done in preparation for such customizability, then AMD would have been much better off designing a straight 8-core CPU for R3/5/7. If AMD knows what it is doing, it already has multiple fabric designs to accommodate different chip designs with more and fewer fabric endpoints (CCX, MP, MCM, IGP, DSP or anything else that might come along in the future) ready to go. Removing a whole CCX should require little more than swapping out the fabric for one that has one less port, deleting everything related to the 2nd CCX and redoing the floor plan. Test-wise, everything is the same as Raven Ridge APUs except for the lack of an IGP, which means very little extra work: if AMD has any foresight, the R3/5/7 test suites are already written to accommodate any foreseeable CCX configuration.

There are many simple ways to drastically reduce ASIC development costs. The most expensive part of removing a CCX should be the new sets of lithography masks as most development quirks will be bugs AMD would have ended up chasing sooner or later regardless of what else it does since everything is built from the same building blocks.

BTW, in the entry-level GPU space, AMD seems happy enough to iterate chips at every ~33% increase (~25% reduction) in die area.
 
AMD doesn't charge $169 in the first place. That's retail price after all in-between parties have added their profit margin.
If the chip cost $42 to manufacture, then the other costs involved to make it a product ready for shipping (PCB with connectors, mounting onto PCB, IHS, adding Stealth cooler, printed paperwork, cardboard box, plastic container, and so on) will surely add $5-10. Then add margin for overhead like marketing, administration and R&D. Wouldn't be surprised if the net production cost of a Ryzen 5 1400 is ~$65.
With a (very low) 30% profit margin to that the factory sale price will be $84.5.
Then you need to (very conservatively) double that to add costs and profit margins for distribution and retail, and end up at a $169 retail price with (too) low profits for everybody involved.

How do you come to that conclusion?
I ordered a Ryzen 5 1600 for myself just the other day. I think it's a fantastic multi-purpose CPU at just the right price point. Compared to a Core i5 at the same price I get now roughly the same (more than sufficient) gaming performance and decidedly better productivity. Comparing these very same CPUs again three years from now I expect the Ryzen to be better in just about every aspect.

That CPU is just silly.

The production and sales of Ryzen has just begun. I don't think defect density is alarming, but sales of four core Ryzen can initially be expected to be so low that it's easier and cheaper to just use the same production line for all wafers while rectifying any teething problems with the production and finish a single CCX design.

With the teething problems gone and sales increasing I expect a single CCX design to emerge. Wouldn't be surprised if that's the Ryzen 3.

Here's a speculation of mine what will be presented in Q3:
Ryzen 5 1450: Single CCX, 4C/8T, 8MB L3, 3.3GHz.
Ryzen 3 family: Single CCX, 4C/4T, up to 8MB L3, various speeds.
 


Yeah, compared to the bargain basement CPUs and APUs AMD were pushing, profit margins should improve. However the desktop market is low margin compared to servers and workstations, while the volumes are pretty low compared to mobile, for example.
But even the 1800X is cheap relative to the transistor count, die size, number of cores, amount of cache and almost any other metric. The 6950X features 3.2 billion transistors and initially retailed for $1700. The harvested, 8-core 6900K version for $1089. The 4.8 billion transistor R7 started at $329 for the 1700, with the 1800X topping out at $499. This was a deliberate move by AMD to shake up the HEDT market and make headlines. They could have sold all 3 models for hundreds more, but they needed the minds and market share of high end desktop users because they were the laughing stock of the industry.
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom